The cryptography of Bitcoin - Amazon S3 · ("mining") Cryptographic ingredients Bitcoin ledger Hash...

transcript

The cryptography of Bitcoin Dr Douglas Stebila School of Mathematical Sciences & School of Electrical Engineering and Computer Science 20 May 2014

http://www.douglas.stebila.ca/research/presentations/

Overview of Bitcoin

What is Bitcoin?

Bitcoin is a decentralized distributed system for establishing a public ledger of transactions.

Basic idea

1.  There's a public ledger that everyone can read with everyone's balance.

2.  Alice wants to pay Bob 3 units. 3.  Alice requests to put a transaction in the

ledger saying "Alice pays Bob 3 units." 4.  The maintainer of the ledger checks

(a) that Alice has big enough balance and (b) that Alice really made the request, then records the transaction in the ledger.

5.  Bob now has a higher balance.

Problems with the basic idea

No anonymity How to verify someone has authorization to spend from Alice's account? Who maintains the ledger?

•  Use public keys rather than names.

•  Use transaction references rather than accounts.

•  Use digital signatures to demonstrate ownership of currency from previous transaction.

•  Distributed ledger: incentivize community to maintain.

Transaction

"Alice pays Bob 3 units." "Alice transfers control of 3 units to Bob."

Input: • Previous transaction ID. • Public key used in previous transaction.

• Digital signature using based on previous transaction's public key.

Output: • Bob's address • # of units

•  Bitcoin address = hash of public key

• Should include own address to "make change"

Transaction

Output: address 320e1d53baf1c

value 2 BTC address f23ea089a76b

value 1.5 BTC

Input: transaction 24d89c02e7ba1 public key

3048c9d000a11789ed signature

9b8d910afa0b0476c

Header +

a list of transactions …

Hash of previous block

Hash of transactions

puzzle difficulty

puzzle solution

Blockchain

A sequence of blocks = ledger of transactions

Which blockchain? Blocks form a tree. • Could have forks in the tree. • Only the longest chain is considered to be valid by the community.

Adding blocks to the chain

A block can only be added to the blockchain if the hash of the block is small. • Users try to generate a block with a small hash. •  ("cryptographic puzzle")

• Updating the blockchain requires work but maintains the public ledger.

• Motivation: whoever constructs the block includes one transaction paying themselves 25 BTC ("mining")

Cryptographic ingredients

Bitcoin ledger

Hash functions

(SHA-256, RIPEMD-160)

Cryptographic puzzles

(Hashcash with SHA-256)

Digital signatures and

public keys (ECDSA)

Hash functions

H : {0,1}* −> {0,1}λ

A public function H that is •  fast and easy to compute •  takes as input arbitrary-length binary strings • outputs a message digest of fixed length

Security properties of hash functions

Collision-resistant

It should be hard to find any two different inputs x1 and x2 such that H(x1) = H(x2).

One-way (preimage resistant)

Given a value y, it should be hard to find any input x such that H(x) = y.

Second-preimage resistant

Given an input x1, it should be hard to find a different input x2 such that H(x1) = H(x2).

Building cryptographic hash functions

Cryptographic hash functions need to take arbitrary-sized input and produce a fixed size output.

Idea: use a fixed-size compression function applied to multiple blocks of the message.

Compression function

h : {0, 1}� ⇥ {0, 1}� ! {0, 1}�

Merkle–Damgård construction •  Break message m into λ-bit blocks m1 || m2 || … || ml•  Add padding. •  Input each block into compression function h along with chained

output; use standardized initialization vector IV to get started.

m1 m2 m3 . . . m`

m1 m2 m3 . . . m` pad

h h h. . .

h hIV H(m)

SHA-256 • Part of the SHA-2 family standardized by NIST in 2001.

• Merkle–Damgård construction.

• Compression function is 64 iterations of function at right.

• No known attacks on SHA-256 (yet) but progress on simplified / reduced-round versions. http://en.wikipedia.org/wiki/SHA-2

Randomness • SHA-256 is not random: it is a deterministic function.

• Does it "look random"?

• How can we tell if a function is random?

http://xkcd.com/221/

Pseudorandomness

"Avalanche effect": changing 1 bit of the input should change around half of the output bits. Golomb's postulates for sequences.

• Assuming SHA-256 is "random" is a stronger assumption than assuming it's collision-resistant / one-way / second-preimage-resistant.

• No known attacks distinguishing SHA-256 from random.

Cryptographic puzzles

Cryptographic puzzle

A "moderately hard" computational task.

Example: • Let H be a hash function with 256 bits of output.

• Find a value x such that H(x) starts with 32 zeros.

Analysis: •  Assume H is a random function

(output bits are independent and identically distributed).

•  Then for each different input x and each i, the probability that the ith bit of H(x) is zero is ½.

•  The probability that the first 32 bits of H(x) are all zero is 1 / 232.

•  Need to try about 231 different x values on average to find a satisfying value. "difficulty"

Hashcash cryptographic puzzle

Example: • Let H be a hash function with λ bits of output. •  Interpret output as an integer between 0 and 2λ−1

• Let s be a string. • Let t be an integer.

• Find a value x such that H(s || x) ≤ t.

Puzzles in Bitcoin

Every miner is trying to construct a block header where H(H(block header || solution)) ≤ difficulty target H = SHA-256 Keep trying random solutions until one works

Hash of previous block

Hash of transactions

puzzle difficulty

puzzle solution

Reward transaction for miner (25 BTC)

Bitcoin mining

Difficulty target adjusted every 2 weeks so that average block generation time is 10 minutes. Current mining rate: • 75.7 quadrillion (approx. 256 hashes) per second • http://blockchain.info/stats, 2014/05/19

Mining pools

Since finding the solution to a new block is so unlikely individually, miners work together in pools. If anyone in the pool finds the solution to the puzzle, the whole pool shares the reward.

How to split the reward? •  Just like Bitcoin mining, but

with a higher difficulty target

•  Pool miners submit whenever they find a hash less than the pool difficulty target

•  Even if it's not a valid Bitcoin block, it still demonstrates that you are working hard

•  Reward split based on number of submitted hashes

scrypt

An alternative cryptographic puzzle used in other cryptocurrencies e.g. Litecoin.

Bitcoin's cryptographic puzzle is computationally bound.

• Easy to run on low memory GPUs or small custom ASICs.

scrypt is memory-bound.

• Needs large amount of memory.

• Won't work well on GPUs.

• Expensive to build custom ASICs.

Digital signatures

Message authentication

How can we be sure Alice really sent a message?

Symmetric message authentication codes: • Alice and Bob share a secret key k

• Alice computes t = MAC(k, m)

• Alice sends (m, t) • Bob checks if t = MAC(k, m)

Problem: how do Alice and Bob share a secret key in the first place?

Problem: How can anyone publicly verify the authentication?

message authentication

•  secret key cryptography

digital signatures

•  public key cryptography

Digital signatures Key generation: Alice generates a pair of related keys: • verification key vk

•  published in a phone book / transaction record

• signing key sk •  kept secret by Alice

Sign(sk, m): Alice uses her signing key sk to generate a signature σ

Verify(vk, m, σ): Anyone can use Alice's verification key vk to check if σ corresponds to m

Security goals of digital signatures

Key recovery

It should be hard compute Alice's signing key sk given just her verification key vk.

Unforgeability

It should be hard to forge a new valid message-signature pair, given Alice's verification key.

•  Forged message doesn't have to be meaningful.

•  Even given copies of other signatures.

•  Even if attacker can choose which messages are signed.

Building a digital signature scheme

modular arithmetic

elliptic curves

groups

ECDSA •  Elliptic

Curve Digital Signature Algorithm

Modular arithmetic

"Clock" arithmetic

Example: 6 o'clock + 8 hours = 14 o'clock = 2 o'clock

Modular arithmetic

m: modulus r = a mod m •  r: the remainder you get

when you divide a by m Example: •  14 mod 12 = 2 •  6 + 8 mod 12 = 2 •  2 × 7 mod 12 = 2

+ 8 hours =

Modular exponentiation

Let g, x, and m be positive integers.

gx mod m represents multiplying g by itself mod m for x times

• Can compute gx mod m efficiently even for very large (500+ digit) values using square-and-multiply algorithm.

Discrete logarithm problem

DLP for mod. exp.

1.  Let g and m be positive integers.

2.  Let x be picked randomly from 0 to m−1.

3.  Compute y = gx mod m.

4.  Given (g, m, y), find x.

Difficulty

Intuitively, DLP for modular exponentiation is hard because mod m makes things wrap around in an "unpredictable" way.

Primitive roots

Exponentiation mod 7 Primitive roots

Notice that some values of g generate all the values from 1 to m−1.

Such g are called generators or primitive roots.

g g2 g3 g4 g5 g6

1 1 1 1 1 1 2 4 1 2 4 1 3 2 6 4 5 1 4 2 1 4 2 1 5 4 6 2 3 1 6 1 6 1 6 1

Abelian groups ("because making things abstract makes them better") An abelian group (G, ×) is a set G and an operation × such that: • × is associative: a × (b × c) = (a × b) × c

• × has an identity 1 such that 1 × a = a = a × 1

• × has inverses: every a has a b such that a × b = 1

• × is commutative: a × b = b × a

A cyclic group of order q is a group G that has a generator g such that g, g2, g3, g4, …, gq−1 is exactly the set of elements of G. Example: •  integers modulo a prime with multiplication are an abelian group

Digital signatures from abelian groups

Let g be the generator of a cyclic group of prime order q. Let H be a hash function. Key generation: • pick x randomly between 0 and q−1

• verification key: vk = gx

• signing key: sk = x

Sign(sk, m): • pick k randomly between 0 and q−1

•  r = gk mod q • s = k−1(H(m)+xr) mod q • signature: σ = (r, s)

Verify(vk, m, σ): • w = s−1 mod q • a = H(m) × w • b = r × w mod q • v = ga × yb mod q • valid if v = r

Attacking the signature scheme

Key generation: •  pick x randomly between 0 and q−1 •  verification key: vk = gx

•  signing key: sk = x Sign(sk, m): •  pick k randomly between 0 and q−1 •  r = gk mod q •  s = k−1(H(m)+xr) mod q •  signature: σ = (r, s) Verify(vk, m, σ): •  w = s−1 mod q •  a = H(m) × w •  b = r × w mod q •  v = ga × yb mod q •  valid if v = r

If you could find x given gx, you could recover the signing key and forge signatures.

•  "discrete logarithm problem"

If you could find m1 ≠ m2 such that H(m1) = H(m2), then you could confuse a signature for m1 as a signature for m2.

•  collision-resistance of H

Digital signatures from abelian groups

DSA: Modular arithmetic

• Group is integers modulo a prime p

•  For high security, need: •  p ≈ 22048 •  public keys are 2048 bits

long •  signatures are 4096 bits long

ECDSA: Elliptic curves

• Group is set of points on a discrete elliptic curve

•  For high security, need: •  256-bit curve •  public keys are 257 bits long •  signatures are 512 bits long

•  ECDSA is faster and has smaller values for same level of security

Elliptic curve

An elliptic curve over the reals is the set of real points (x, y) satisfying an equation of the form

y2 = x3 + ax + b for fixed real numbers a and b.

Elliptic curve points as a group • G = set of points on the curve • operation = "point addition" • can make equations from following geometric intuition

Elliptic curve scalar-point multiplication

• Let P be a point on the curve. • kP represents adding P to itself k times.

Multiplicative groups Additive groups multiplication: g × h addition: P + Q squaring: g2 doubling: 2P exponentiation: gx scalar-point multiplication: kP square-and-multiply algorithm double-and-add algorithm

Discrete logarithm problem

Multiplicative groups

1.  Let g be a generator of a cyclic group of prime order q.

2.  Let x be picked randomly from 0 to q−1.

3.  Compute y = gx.

4.  Given (g, q, y), find x.

Additive groups

1.  Let P be a generator of a cyclic group of prime order q.

2.  Let k be picked randomly from 0 to q−1.

3.  Compute Q = kP.

4.  Given (P, q, Q), find k.

Difficulty of DLP

Best known algorithm for DLP that works in every group:

• Pollard's rho algorithm ≈ sqrt(q) operations

The properties of some groups make it easier. • mod p: number field sieve •  p ≈ 22048 & q ≈ 2210

=> 105-bit security

Elliptic curve groups: • nothing better than sqrt(q) •  q ≈ 2256

=> 128-bit security

Elliptic curves over prime fields

Use modular arithmetic instead of real numbers: y2 = x3 + ax + b (mod p)

http://arstechnica.com/security/2013/10/a-relatively-easy-to-understand-primer-on-elliptic-curve-cryptography/2/

3 � x+ 1 (mod 97)

Digital signatures in Bitcoin

Elliptic curve digital signature algorithm using the NIST p256 elliptic curve group.

• mod p ≈ 2256

Best known algorithm for forging signatures takes about 2128 operations.

•  ≈ 248 ≈ 1024 years for 10 million 4GHz computers

•  universe is ≈ 1010 years old

Bitcoin transaction

Output:

address 320eed53baf1c

value 2 BTC address f23eed89a76b

value 1.5 BTC

Input: transaction 24d89c02e7ba1

public key 3048c9d000a11789ed

signature 9b8d910aaa0b0476c

Public key: • ECDSA public verification

key used in address from previous transaction

Signature: •  signature of transaction

using corresponding ECDSA private signing key

Bitcoin address: RIPEMD-160( SHA-256(ECDSA public key)

Cryptographic parts of Bitcoin ledger

Transactions Blockchain

Hash used to chain transactions together (SHA-256) Only blocks in longest chain considered valid

Blocks

Hash used to collect transactions together Cryptographic hash puzzle required to make block valid (Hashcash SHA-256)

Digital signatures for transaction approval (ECDSA) Hashed public keys for addresses

Breaking Bitcoin via cryptography

Forge transactions

Breaking elliptic curve discrete logarithm with classical computers needs mathematical breakthrough.

•  Become a mathematical supergenius.

Quantum computers can easily break ECDLP.

•  "Just" need to build a quantum computer.

Mine faster

Figure out how to break partial preimage resistance / pseudorandomness of SHA-256.

•  Would break lots of other stuff on the Internet.

The cryptography of Bitcoin - Amazon S3 · ("mining") Cryptographic ingredients Bitcoin ledger Hash...

Documents